Research On Deep Learning Model Of Abnormal Gait In Parkinson’s Disease

Parkinson’s disease is the second-largest neurodegenerative disease,and its prevalence is currently on the rise.There is no clinical cure for Parkinson’s disease.It is necessary to diagnose patients’ clinical symptoms early,monitor disease progression,and quickly determine the best treatment plan.At present,the diagnosis and classification of Parkinson’s disease are clinically subjective,which will cause delays and increase the burden on families and society of patients with Parkinson’s disease.Therefore,there is an urgent need for objective diagnosis and quantitative classification of Parkinson’s disease.This article takes the abnormal gait of Parkinson’s disease as the research object,from the aspects of feature extraction and feature construction,and makes full use of deep learning models and statistical analysis to carry out objective diagnosis and quantitative grading research of abnormal gait of Parkinson’s disease.Main work:(1)Introduce deep learning to automatically extract the nonlinear abnormal gait features embedded in the gait time series and establish a model for distinguishing abnormal gait classification of Parkinson’s disease with superior generalization performance;(2)Build a new accurate representation of Parkinson’s disease The gait symmetry quantitative index of abnormal gait characteristics of the disease is helpful to assist clinical diagnosis.The innovation results are as follows:(1)Propose multi-sensor-oriented multi-channel convolutional neural network deep learning recognition model for human actions.The model introduces a multi-channel convolution structure,which integrates a multi-layer convolution layer and a multi-channel convolution structure,fully utilizes the multi-scale convolution kernels of different lengths to capture time-series feature characteristics at multiple scales,and extracts the most representative features that are closely related to changes in human movement non-linear gait features are helpful to improve the accuracy of action recognition.Experimental results show that the accuracy of action recognition of the model proposed in this paper can reach96.03%.Compared with deep learning models such as LSTM,it can better control the balance between high accuracy and low computational complexity,and effectively improve the performance of human action recognition.(2)Propose a Parkinson’s abnormal gait identification model based on multi-channel dilated convolutional neural network.In view of the high-frequency characteristics of Parkinson’s abnormal gait signals,the model introduces an expanded convolution structure and a hidden fully connected layer,making full use of the expanded convolution to dynamically increase the fusion ability of the convolution kernel receptive field and the fully connected layer for features,and efficient extraction The non-linear characteristic information of Parkinson’s abnormal gait contained in the gait signal can effectively improve the accuracy of identifying abnormal gait.The experimental results show that the model proposed in this paper can accurately extract nonlinear abnormal feature information closely related to Parkinson’s abnormal gait,and the recognition accuracy is 99.21%,which significantly improves the discrimination rate of Parkinson’s abnormal gait of different severity,and is expected to be that are auxiliary diagnostic tools for Parkinson’s disease grading.(3)Propose a new "area symmetry index" that characterizes the symmetry of Parkinson’s abnormal gait.This symmetry index characterizes the gait difference between the left and right lower limbs by the area between the pair of left and right lower limb stride sequences,thereby quantifying the gait symmetry of the human body.The experimental results showed that the area symmetry indexes of healthy control group,mild Parkinson’s disease group and severe Parkinson’s disease group were(2.08 ± 0.41),(3.09 ± 0.96)and(4.75 ± 2.37),and the differences between the groups were significant(p <0.05),indicating that the area symmetry index can effectively quantify the symmetry of human gait.Further statistical analysis found that the area symmetry index has strong correlation with Parkinson’s disease H&Y level,and the correlation coefficient is 0.832,which is expected to become an important feature of the quantitative classification of Parkinson’s disease.